I was doing some simulations in R and I noticed I was missing exactly one simulation, after looking into my code I discovered this strange R behaviour:
N <- 10000
prop <- 0.9
M <- N * (1 - prop)
print(M)
[1] 1000
print(as.integer(M))
[1] 999
Asked
Active
Viewed 179 times
2
ArturoSaCo
- 319
- 1
- 10
-
Looks like floating point issue. – CHP Apr 11 '13 at 10:01
-
1..and `as.integer()` rounds _toward zero_. – Anthony Damico Apr 11 '13 at 11:46
-
My confusion had to do more with the rounding than with the non-equality NPE's answer clarified that. – ArturoSaCo Apr 11 '13 at 15:01
1 Answers
4
The issue is that 0.9 cannot be represented exactly as a binary floating-point value:
> sprintf('%.20f', 0.9)
[1] "0.90000000000000002220"
All the calculations that follow give inexact results:
> sprintf('%.20f', 10000 * (1 - 0.9))
[1] "999.99999999999977262632"
NPE
- 486,780
- 108
- 951
- 1,012